HELIX: Scaling Raw Audio Understanding with Hybrid Mamba-Attention Beyond the Quadratic Limit

While self-supervised learning (SSL) has revolutionized audio representation, the excessive parameterization and quadratic computational cost of standard Transformers limit their deployment on resource-constrained devices. To address this bottleneck, we propose HEAR (Human-inspired Efficient Audio Representation), a novel decoupled architecture. Inspired by the human cognitive ability to isolate local acoustic features from global context, HEAR splits the processing pipeline into two dedicated modules: an Acoustic Model for local feature extraction and a Task Model for global semantic integration. Coupled with an Acoustic Tokenizer trained via knowledge distillation, our approach enables robust Masked Audio Modeling (MAM). Extensive experiments demonstrate that HEAR requires only 15M parameters and 9.47 GFLOPs for inference, operating at a fraction of the computational cost of conventional foundation models (which typically require 85M-94M parameters). Despite this high efficiency, HEAR achieves highly competitive performance across diverse audio classification benchmarks. The code and pre-trained models are available at https://github.com/HarunoriKawano/HEAR

Fine-tuning Transformer-based foundation models has become the dominant strategy for domain adaptation in audio and speech processing. To reduce the computational and memory costs of this process, parameter-efficient transfer learning (PETL) methods have been widely explored. Meanwhile, Mamba, a recent state-space model, has emerged as a promising alternative to Transformers for sequence modeling. In this work, we present MambAdapter, a parameter-efficient transfer learning approach that integrates Mamba into low-rank bottleneck adapters. Our design combines parameter sharing across adapters with the injection of a lightweight Mamba module, enabling more effective modeling of audio features. We demonstrate that MambAdapter matches or outperforms strong PETL baselines on four audio classification tasks and five speech recognition languages, even when operating under reduced parameter budgets.

A large audio language model (LALM) turns a minute of speech into 750-1,500 tokens and prefills every one. Image-token pruning often cuts after the language model's first layers, where image tokens draw little attention. Audio tokens draw much more attention there, and their ranking is still far from final, so audio needs a ranking before the language model runs. Surprisingly, the attention an audio token will receive across the language model is already linearly predictable from its encoder output, before the language model runs. A linear map, fitted in closed form without labels, predicts this all-layer attention ranking at $ρ\geq .69$ on eleven of thirteen LALMs. Our method, Triage, cuts audio tokens by this prediction and, on multiple choice, cuts again at layer 2, correcting the prediction with the attention observed there. Triage sets its compression without labels, under two budgets that limit how far its output may differ from the model's own full-audio output. At the conservative budget, its word error rate and accuracy stay within .04 of full audio. At the aggressive budget, Triage beats every baseline in all twelve transcription cases. On multiple choice, at 2.2-5x compression, it outperforms DART, the strongest baseline on average, by .043 in mean accuracy. Because it cuts before the language model, it raises the audio that fits in Qwen2.5-Omni-3B's context window from 21.8 to about 62 minutes. At its most compressive point, Triage lets one GPU serve 4x as many concurrent 5-minute streams of that model. Project page: https://audio-triage.github.io